I've added an Aptina MT9P031 camera to an OMAP3530-based board (LogicPD Torpedo), and it's mostly working well except that alternating frames of video output have bad color information (far too much green). I get one frame with the correct colors, then one frame with too much green, and so on. When I turn off the CFA interpolation in the previewer, the output looks good, albeit in black and white. And, oddly, when I use any one of the other three possible Bayer patterns, the colors are stable between frames, though not quite right. It's only when I use the Bayer filter that actually corresponds to my camera that I see this problem.
It looks to me as though the ISP driver in the kernel I have (based on the 2.6.32 mainline) has a single hard-coded Bayer pattern and chops a line or column off the sensor input to have the sensor input conform to the expected Bayer pattern. I suspect that losing a line or column from the input is causing my problem, since that seems to be the only effect of different Bayer patterns in the ISP driver.
Is it possible to change the Bayer pattern in the ISP driver? Changing CCDC_COLPTN doesn't seem to have any effect. I expect the actual CFA interpolation is controlled by the CFA coefficients, which, as I tenuously understand it, control a TI-proprietary algorithm.
Is there something else that could be causing this problem? I think I've ruled out interlacing, and it's suspicious that simply changing the pixel format in the camera driver to a different Bayer pattern makes the alternating frame problem go away.
Thanks,
Rex